Efficient structural pseudoinverse learning-based hierarchical representation learning for skin lesion classification

The success of deep learning in skin lesion classification mainly depends on the ultra-deep neural network and the significantly large training data set. Deep learning training is usually time-consuming, and large datasets with labels are hard to obtain, especially skin lesion images. Although pre-training and data augmentation can alleviate these issues, there are still some problems: (1) the data domain is not consistent, resulting in the slow convergence; and (2) low robustness to confusing skin lesions. To solve these problems, we propose an efficient structural pseudoinverse learning-based hierarchical representation learning method. Preliminary feature extraction, shallow network feature extraction and deep learning feature extraction are carried out respectively before the classification of skin lesion images. Gabor filter and pre-trained deep convolutional neural network are used for preliminary feature extraction. The structural pseudoinverse learning (S-PIL) algorithm is used to extract the shallow features. Then, S-PIL preliminarily identifies the skin lesion images that are difficult to be classified to form a new training set for deep learning feature extraction. Through the hierarchical representation learning, we analyze the features of skin lesion images layer by layer to improve the final classification. Our method not only avoid the slow convergence caused by inconsistency of data domain but also enhances the training of confusing examples. Without using additional data, our approach outperforms existing methods in the ISIC 2017 and ISIC 2018 datasets.

Detection Method of Computer Room Personnel Based on Improved DERT

Recently developed methods such as DETR [1] apply Transformer [2] structure to target detection. The performance of using Transformers for target detection (DETR) is similar to that of two-stage target detector. First of all, this paper attempts to apply Transformer to computer room personnel detection. The contributions of the improved DETR include: 1) in order to improve the poor performance of small target detection. Embed Depthwise Convolution in the encoder. When the coding feature is reconstructed, the channel information is retained. 2) in order to solve the problem of slow convergence in DETR training. This paper improves the cross-attention in DECODE and adds the spatial query module. It can accelerate the convergence of DETR. The convergence speed of the improved method is six times faster than that of the original DETR, and the mAP0.5 is improved by 3.1%.

Bacterial Foraging Algorithm With Potential Field Guidance Mechanism

To address the problems of weak quorum sensing ability and slow convergence speed in bacterial foraging algorithm, a bacterial foraging algorithm with potential field guidance mechanism is proposed. The algorithm combines the sampling guidance mechanism in the artificial potential field algorithm to provide the optimization direction for each bacterium; The original swimming operation of bacterial foraging algorithm is used to realize the local optimization strategy, and the local dimension update is added after swimming, so that the search range of bacteria in chemotaxis operation is wider; In the elimination and dispersal operation of bacterial foraging algorithm, double Gaussian function is introduced to re initialize the location of bacteria, so as to better avoid the algorithm falling into local extremum and improve the optimization ability of the algorithm. The experimental results show that the improved bacterial foraging algorithm has better optimization ability than the basic bacterial foraging algorithm.

Application study of UAV path planning based on the balanced search factor artificial bee colony algorithm

In this paper, aiming at the shortcomings of slow convergence speed and weak local search ability of traditional artificial bee colony algorithm in path planning, an artificial bee colony algorithm based on balanced search factor is proposed for UAV path planning. Using a search strategy based on balanced search factor, the depth search is carried out while maintaining a certain population diversity. The global search ability and local development ability are balanced, the average accuracy of path planning is improved, the robustness of path planning is enhanced, and the ability to obtain better path solutions is improved.

Solving the hourglass instability problem using rare mesh variation-difference schemes

The hourglass instability effect is characteristic of the Wilkins explicit difference scheme or similar schemes based on two-dimensional 4-node or three-dimensional 8-node finite elements with one integration point in the element. The hourglass effect is absent in schemes with cells in the form of simplexes (triangles in two-dimensional case, tetrahedrons in three-dimensional case). But they have another well-known drawback - slow convergence. One of the authors proposed a rare mesh scheme, in which elements in the form of a tetrahedron are located one at a time in the centers of the cells of a hexahedral grid. This scheme showed the absence “hourglass” effect and other drawbacks with high efficiency. This approach was further developed for solving 2D and 3D problems.

Automatic Film Label Acquisition Method Based on Improved Neural Networks Optimized by Mutation Ant Colony Algorithm

Nowadays, with the constant change of public aesthetic standards, a large number of new types and themes of film programs have emerged. For this reason, this paper proposes an improved neural network optimized by mutation ant colony algorithm for automatic acquisition of film labels, which not only overcomes the disadvantages of traditional neural network, such as difficulty in determining weights, slow convergence speed, and easiness to fall into local minimum, but also makes up for the shortcomings faced by using ant colony algorithm alone through the gradient information of quantum genetic algorithm neural network. The results show that the user similarity judgment is added in the process of calculating the user rating deviation between movies, and the neighbor chooses to add the movie tag weight and rating similarity as the basis for the neighbor selection of the target movie in the process of predicting the target movie rating. Experiments show the effectiveness of the algorithm.

Improved Memetic NSGA-II Using a Deep Neighborhood Search

Elitist non-sorted genetic algorithms as part of Pareto-based multi-objective evolutionary algorithms seems to be one of the most efficient algorithms for multi-objective optimization. However, it has some shortcomings, such as low convergence accuracy, uneven Pareto front distribution, and slow convergence. A number of review papers using memetic technique to improve NSGA-II have been published. Hence, it is imperative to improve memetic NSGA-II by increasing its solving accuracy. In this paper, an improved memetic NSGA-II, called deep memetic non-sorted genetic algorithm (DM-NSGA-II), is proposed, aiming to obtain more non-dominated solutions uniformly distributed and better converged near the true Pareto-optimal front. The proposed algorithm combines the advantages of both exact and heuristic approaches. The effectiveness of DM-NSGA-II is validated using well-known instances taken from the standard literature on multi-objective knapsack problem. As will be shown, the performance of the proposed algorithm is demonstrated by comparing it with M-NSGA-II using hypervolume metric.

Prediction of Enterprise Financial Crisis based on Improved Neural Network

Objectives: Neural network is a very important research model in human brain research, and it has been cross researched and applied in many disciplines and fields. Methods: However, there are some shortcomings in the neural network, such as long learning cycle and slow convergence speed. Results: Therefore, in this paper, the enterprise financial crisis prediction based on improved neural network was proposed. Then in the light of the shortcomings of the neural network, the optimization and improvement were carried out. After that, the genetic algorithm was introduced to update the neural network structure and improve the prediction accuracy of the neural network. Finally, the improved neural network was applied to the financial crisis prediction, and good results were achieved. Conclusion: It is proved that the research has good application value and promotion prospect.

Research on adaptive non-singular fast terminal sliding mode control based on variable exponential power reaching law in manipulators

Aiming at the manipulator control system is susceptible to model parameter uncertainty and external disturbance. In this article, an adaptive non-singular fast terminal sliding mode control based on variable exponential power reaching law is proposed. First, due to the slow convergence speed and large chattering of the traditional reaching law, the variable exponential power reaching law is designed in this article. It can adaptively change the reached speed according to the system state, improve the accuracy of the control system and reduce chattering. Second, compared to the slow convergence speed of traditional sliding mode surfaces, this article uses non-singular fast terminal sliding mode surfaces to speed up the system error convergence speed. At the same time, in view of the problem that the disturbance has an uncertain upper bound in the actual problem, the adaptive law is used to estimate the uncertain upper bound of the system disturbance. And by introducing a time-varying boundary layer to improve the symbolic function in the control law. Finally, the Lyapunov function is used to prove the stability of the control system. The simulation results show that the controller designed in this article has good position tracking performance and strong anti-disturbance ability.

Reverse guidance butterfly optimization algorithm integrated with information cross-sharing

Aiming at the disadvantages of slow convergence and the premature phenomenon of the butterfly optimization algorithm (BOA), this paper proposes a modified BOA (MBOA) called reverse guidance butterfly optimization algorithm integrated with information cross-sharing. First, the quasi-opposition concept is employed in the global search phase that lacks local exploitation capabilities to broaden the search space. Second, the neighborhood search weight factor is added in the local search stage to balance exploration and exploitation. Finally, the information cross-sharing mechanism is introduced to enhance the ability of the algorithm to jump out of the local optima. The proposed MBOA is tested in fourteen benchmark functions and three constrained engineering problems. The series of experimental results indicate that MBOA shows better performance in terms of convergence speed, convergence accuracy, stability as well as robustness.